[NEW] 2025:Hallucinations in Gen-AI

Explore the foundations of artificial intelligence, machine learning, and deep learning, and see how transformers and large language models enable generative ai to create text, images, and code.

Learn to use generative ai as a thought partner, reasoning tool, and writing assistant to brainstorm, outline, expand, and summarize, while avoiding fact retrieval, to improve LMS workflows.

Master prompt engineering by crafting clear prompts that yield accurate, relevant results from language models, with Llama 2, and comparisons to OpenAI and Gemini.

Generative AI writing use cases harness brainstorming, poetry, emails, and recipes. Create prompts for sales, translations, and information retrieval from file base or database to ensure accuracy and avoid hallucinations.

Explore Gen AI reading use cases, from proofreading and editing to summarizing articles, emails, and multilingual conversations, and tailor outputs for legal, marketing, or other domain-specific insights.

Explore Gen AI chat use cases from internal bots aiding agents to bot triage and live chats, addressing hallucinations and differentiating general versus specialized legal, medical, and travel chatbots.

Explain what hallucination is in gen-ai, show how models fabricate when lacking data, and empower users to validate facts, quote unknowns, and use fact sheets to avoid fabrication.

Identify what leads to hallucination in language models, including data gaps, overfitting, architecture and training choices, prompt complexity, randomness, and evaluation metrics.

Identify and mitigate hallucinations through root cause analysis, inconsistency checks against authoritative sources, plausible and error analysis, and human in the loop feedback with transparent high quality data practices.

Evaluate hallucination in gen-ai models by applying rouge scores, blue scores, rouge clipping, and n-gram based measures (unigram to LCS) to judge text fidelity.

Explore how the Bleu score evaluates machine translation quality by comparing generated translations with human reference translations, highlighting word-level matches and the score's speed and cost-effectiveness.

Learn to improve LLM results by clearly defining tasks, iteratively refining prompts, and controlling the environment, knowledge cutoff, and output length to minimize hallucinations and bias.